---
library_name: openpi
tags:
  - openpi
  - pi0.5
  - lora
  - lerobot
  - robotics
  - sim2real
  - mug-into-bowl
---

# OpenPI pi0.5 LoRA for sim2real mug into bowl

This repository stores the final OpenPI checkpoint for the `pi05_clutter_libero_lora` SFT run.

## Dataset

This model was SFT-trained using the `IDEAS-Lab-Northwestern/sim2real-mug-into-bowl` dataset:

https://huggingface.co/datasets/IDEAS-Lab-Northwestern/sim2real-mug-into-bowl

## Checkpoint

The uploaded checkpoint preserves the OpenPI checkpoint directory layout:

```text
checkpoints/
  pi05_clutter_libero_lora/
    sim2real_mug_from_sim_table_25000_30k/
      30000/
        _CHECKPOINT_METADATA
        params/
        train_state/
        assets/
```

Use the checkpoint directory with OpenPI as:

```text
checkpoints/pi05_clutter_libero_lora/sim2real_mug_from_sim_table_25000_30k/30000
```

## Training Summary

- Config: `pi05_clutter_libero_lora`
- Experiment: `sim2real_mug_from_sim_table_25000_30k`
- Dataset: [`IDEAS-Lab-Northwestern/sim2real-mug-into-bowl`](https://huggingface.co/datasets/IDEAS-Lab-Northwestern/sim2real-mug-into-bowl)
- Resume source: local `sim_table_lora/25000` checkpoint
- Final step: `30000`
- Batch size: `4`
- Base model family: OpenPI pi0.5 LoRA